The accuracy and reproducibility of magnetic resonance (MR) imaging in the determination of left ventricular mass in humans was investigated. Left ventricular wall volume was measured from ten short-axis, end-diastolic MR images that spanned the left ventricle. Mass was estimated on the basis of average left ventricular wall volume and an assumed myocardial density. To establish the accuracy of the technique, the authors imaged ten cadaver hearts and compared true left ventricular weight with the mass estimate based on MR imaging findings. In vivo determination of left ventricular mass was evaluated in 40 subjects, with resultant calculated masses of 156.4-319.3 g. Intra- and interobserver variabilities of the technique were analyzed in ten subjects. Both the intra- (r = .96, standard error of estimate [SEE] = 11.1 g) and interobserver variabilities (r = .91, SEE = 17.8 g) were excellent. Eight subjects were imaged on two separate occasions to evaluate reproducibility of the technique and confidence limits for a given measurement. For these eight, there was good correlation between the two estimates (r = .93, SEE = 21 g). The authors conclude that MR imaging yields highly accurate and reproducible estimates of left ventricular mass in humans in vivo.